Feature extraction

Abstract

In the previous chapters we have examined static and dynamic methods of program analysis. These features must be translated into mathematical representations and birthmarks to be useful. Furthermore, mathematical representations may be embedded in other mathematical types to make birthmarks more amenable to similarity comparisons and for use in classification algorithms. Another approach is to represent features using kernels. This allows for the use of classification algorithms including the support vector machine for complex data types. This chapter examines the mathematical representations that we use to describe program features.